Summary: Researchers have bridged the gap between biology and silicon by creating a 3D programmable device that merges living brain cells with advanced electronics. Unlike previous “brain-on-a-chip” attempts that grew cells on flat surfaces, this device uses a flexible, microscopic metal mesh as a scaffold, allowing tens of thousands of neurons to grow around and through the sensors.
The study demonstrates that this “biological neural network” can be trained to recognize complex electrical patterns, offering a high-efficiency alternative to power-hungry AI.
Key Facts
- From the Inside Out: The device uses a 3D mesh of microscopic wires and electrodes supported by a flexible epoxy coating, allowing it to interface directly with the soft tissue of living neurons.
- Long-Term Stability: The team successfully tracked and stimulated the network for over six months, observing how connections between neurons evolved and strengthened over time.
- Pattern Recognition: By training an algorithm to interpret the neurons’ activity, the system correctly distinguished between different spatial and temporal patterns of electrical pulses.
- Energy Efficiency: While modern AI consumes massive amounts of power, the human brain performs similar computations using roughly one-millionth of the energy, a gap this bio-hybrid technology aims to close.
Source: Princeton University
Princeton researchers have combined brain cells and advanced electronics into a single 3D device that can be programmed to recognize patterns using computational techniques.
Past attempts at using brain cells to do computation have relied on 2D cultures grown in a petri dish or 3D clusters that are probed and monitored from outside. The Princeton device takes a different approach, working from the inside out.
Using advanced fabrication techniques, the team created a 3D mesh made of microscopic metal wires and electrodes supported by a thin epoxy coating. Because the coating is so thin, it has just the right amount of flexibility to interface with the soft neurons that grow around it. The team used the mesh as a scaffold to culture tens of thousands of neurons into a vast 3D network that can be used to do computation.
The study was published in Nature Electronics on Apr. 23.
The researchers said the new integrated approach enabled them to record and stimulate the neurons’ electrical activity at a much finer scale than past approaches. They tracked the evolution of the system over a period of more than six months, experimenting with ways to strengthen and weaken connections between key neurons, and ultimately trained an algorithm that could recognize patterns of electrical pulses.
In one test, they used pairs of distinct spatial patterns. In another, they used distinct temporal patterns. The system correctly distinguished among the patterns in both tests. The researchers said they hope to scale the system to the point where it can do increasingly complex tasks.
The work was led jointly by Tian-Ming Fu, assistant professor of Electrical and Computer Engineering and Omenn-Darling Bioengineering Institute; James Sturm, Stephen R. Forrest Professor of Electrical and Computer Engineering; and Kumar Mritunjay, a postdoctoral researcher in electrical and computer engineering.
While initially developed to study fundamental problems in neuroscience, the team realized it could shed light on a key bottleneck of modern AI technology: energy consumption.
“The real bottleneck for AI in the near future is energy,” said Fu. “Our brain consumes only a tiny fraction — about one millionth — of the power consumed by today’s AI systems to perform similar tasks.”
Mritunjay, the paper’s first author, said that systems like this, called 3D biological neural networks, “not only help uncover the computing secrets of the brain but can also assist in understanding and possibly treating neurological diseases.”
Key Questions Answered:
A: In a sense, yes. It is a biological neural network (3D-BNN). By using living neurons as the “processors” and a metal mesh as the “wiring,” the researchers have created a hybrid system that can be programmed to perform specific tasks, like recognizing patterns.
A: The researchers used electrical stimulation to strengthen or weaken the connections between specific neurons, a process called synaptic plasticity. They then used an algorithm to “read” the resulting electrical signals, much like how a computer reads data from a hard drive.
A: That is the long-term goal. Because the mesh is flexible and mimics the brain’s natural structure, it could eventually lead to sophisticated implants that “talk” to the brain in its own language, potentially bypassing damaged areas in patients with neurological disorders.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by our staff.
About this neurotech research news
Author: Scott Lyon
Source: Princeton University
Contact: Scott Lyon – Princeton University
Image: The image is credited to Neuroscience News
Original Research: Closed access.
“A three-dimensional micro-instrumented neural network device” byKumar Mritunjay, James C. Sturm & Tian-Ming Fu. Nature Electronics
DOI:10.1038/s41928-026-01608-1
Abstract
A three-dimensional micro-instrumented neural network device
Three-dimensional (3D) cultured neural networks that emulate the structures and computational principles of the brain could be of use in the development of brain-inspired computing and artificial intelligence, as well as in the understanding of neural development and disease progression.
However, creating such stable device–neural network interfaces remains challenging, limiting the potential of such 3D neural networks. Here we report a 3D micro-instrumented neural network device in which a 3D flexible electronic sensor and stimulator array is integrated with a 3D cultured neural network.
Our device can be used to record action potentials from multiple planes over a period of 6 months, allowing the quantitative monitoring of the evolving connectivity maps and the pharmacological stimulation responses of the neural networks.
This approach also supports chronic electrical stimulation, which we use to train neural networks by tuning the connectivity strengths between neurons, creating a reservoir neural network for biocomputing.
